It is kind of the case, but it does not need to expand the whole node to find the maximum. It samples some children instead from a NN (the Monte Carlo aspect)